QSST: A Quranic Semantic Search Tool based on word embedding
Ensaf Hussein Mohamed, Eyad Mohamed Shokry
Abstract
Retrieving information from the Quran is an important field for Quran scholars and Arabic researchers. There are two types of Quran searching techniques: semantic or concept-based and keyword-based. Concept-based search is a challenging task, especially in a complex corpus such as Quran. This paper presents a concept-based searching tool (QSST) for the Holy Quran. It consists of four phases. In the first phase, the Quran dataset is built by manually annotating Quran verses based on the ontology of Mushaf Al-Tajweed. The second phase is word Embedding, this phase generates features’ vectors for words by training a Continuous Bag of Words (CBOW) architecture on large Quranic and Classic Arabic corpus. The third phase includes calculating the features’ vectors of both input query and Quranic topics. Finally, retrieving the most relevant verses by computing the cosine similarity between both topic and query vectors. The performance of the proposed QSST is measured by comparing results against Mushaf Al-Tajweed. Then, precision, recall, and F-score are computed and their percentages were 76.91%, 72.23% 69.28% respectively. In addition, the results are evaluated by three Islamic experts and the average precision was 91.95%. Finally, QSST results are compared with the recent existing tools; QSST outperformed them.